• Aucun résultat trouvé

LetS >0, T >0, U =℄0;T[℄0;S[ and D=T d =(R=Z) d theattorus

N/A
N/A
Protected

Academic year: 2022

Partager "LetS >0, T >0, U =℄0;T[℄0;S[ and D=T d =(R=Z) d theattorus"

Copied!
17
0
0

Texte intégral

(1)

Yann Brenier

Abstrat. Theoneptofharmoni(orwave)mapsuptorearrangementis

introdued.Arelationisestablishedwiththesmoothsolutionsoftheisother-

malirrotationalinvisidgasdynamisequations.

Onintroduitlanotion d'appliationharmonique a rearrangementpres. On

etablit un lien ave les solutions regulieres des equations derivant la dy-

namiqueisothermedesgazsansvisositenitourbillon.

1. Review of somelassialonepts

1.1. Harmoni (or wave) maps. LetS >0, T >0, U =℄0;T[℄0;S[ and

D=T d

=(R=Z) d

theattorus. Amap(t;s)2U !X(t;s)2D isusuallyalled

aharmonimap(resp. awavemap)ifitisaritialpointofthefuntional

Z

U 1

2 (j

t X(t;s)j

2

+j

s X(t;s)j

2

)dtds;

(1.1)

where =1(resp. = 1), with respetto perturbationswithompat support

in U. (Herej:j denotestheEulidean normonR d

.) ThismeansthatX solvesthe

homogeneousLaplae(resp. wave)equation

tt X+

ss X =0:

(1.2)

Intheharmoniase=1,amapX isalledminimizingharmoniifitminimizes

(1.1)asitsvalueXj

U

alongtheboundaryisxed.

Remark. A natural lass for the \target" D of a harmoni (or wave) map is

thelass of generalRiemannian manifolds. TheaseD =T d

is fairlytrivial but

suÆient for our disussion. Compat Riemmanian manifolds without boundary

ould also be onsidered. However, ompat manifolds with boundaries or non

ompatmanifolds,inpartiulartheEulideanspaeR d

,wouldausediÆulties.

1991MathematisSubjetClassiation. Primary58E2058E3076N;Seondary35J35Q35.

Keywordsandphrases. Harmoniity,gasdynamis,rearrangement,vibratingstrings.

WorkpartlysupportedbytheaustrianprojetSTART(FWF-TEC-Y-137).

(2)

1.2. Laws,rearrangements, Moser'slemma. Let(A;da)beaprobability

spae (typially A = [0;1℄or A = T d

equipped with the Lebesgue measure da).

Forameasurablefuntion a2A!X(a)2D,wedenethe\law"ofX to be

(x)= Z

A

Æ(x X(a))da;

(1.3)

whihisaprobabilitymeasureonD, morepreiselydened by

Z

D

h(x)d(x)= Z

A

h(X(a))da;

(1.4)

forallh2C(D). (Anusualdenominationforis\push-forward"ofdabyX.) We

nowsaythattwosuhmeasurablefuntionsX andY areequaluptorearrangement

iftheyhavethesamelaw,namely

Z

A

Æ(x Y(a))da= Z

A

Æ(x X(a))da:

(1.5)

Letus quoteaveryusefulresult,due to MoserandlaterimprovedbyDaorogna

andMoser[DM℄,atleastinthespeialaseA=T d

Lemma 1.1. Let A =D =T d

. Let (x)> 0be a smooth funtion of x 2D

withmean1. ThenthereisasmoothorientationpreservingdieomorphismX from

A to D suhthat (x)dx isthe law of X. In addition, as depends smoothly on

someparameters2[0;1℄, wemaynd X thatalso depends smoothly ons.

Anelementaryproofisprovidedin setion8.8

1.3. Isothermal irrotational gas dynamis. The evolutionof aninvisid

gas moving in the Eulidean spae R d

(physially d = 1;2;3) is desribed by a

densityeld(s;x)>0,apressureeldp(s;x)>0,atemperatureeld(s;x)>0,

andaveloityeldu(t;x),valuedin R d

,wheresstandsforthetimevariableand

xforthespaevariable. Theseeldsaresubjettothefollowingequations(where

weusenotationsr for(

x

1

;::::;

x

d

)and: fortheEulideaninnerprodutinR d

)

s

+r:(u)=0;

(1.6)

whih expresses the onservation of mass, and is usually alled \the ontinuity

equation",

s

u+(u:r)u+ rp

=0;

(1.7)

whih is equivalent to the onservation of momentum. The evolution is alled

isothermalifthetemperatureeld isaonstant(s;x)=>0andthepressure

isgivenby

p=;

(1.8)

forsomeonstant>0. Introduing

= p

log; (1.9)

theseequationsanbewrittenasasymmetrirst-orderhyperbolisysteminspae

timevariables(x;s)

s

+u:r+ p

r:u=0;

(1.10)

s

u+(u:r)u+ p

r=0;

(1.11)

(3)

forwhihexisteneanduniquenessofsmoothsolutions,forsmoothinitialdataand

shorttime intervals,arestandardresults(see[Si℄forinstane). Potentialveloity

eldsofform

u=r;

(1.12)

with(salar)potentials(s;x)arepreservedduring theevolution,aslongasthey

staysmooth. Then,theontinuityequationbeomes

s

+r:(r)=0 (1.13)

andthemomentumequationanbeintegratedout,whihleads(uptoanirrelevant

onstant)to:

s +

1

2 jrj

2

+log=0:

(1.14)

Then,equations (1.13),(1.14), from aself-onsistentsystemofevolutionPDEsin

and,desribingisothermalirrotationalinvisidgasdynamis.

Remarkonthe Ationpriniple. Ataformallevel,equations(1.13),(1.14)an

beeasilyderivedbyvaryingtheAtion

Z

( jEj

2

2

log)dsdx;

(1.15)

withrespetto (s;x)>0,E(s;x)2R d

, subjetto

s

+r:E=0:

(1.16)

Indeed, (s;x) is introdued as a Lagrange multiplier for onstraint (1.16). By

varyingtheassoiatedLagrangian

Z

( jEj

2

2

log

s

r:E)dsdx;

with respet to (;E;), we get E = r and (1.14). Of ourse, as usual for

hyperboli equations,the Ationprinipledoesnotorrespond toaminimization

problem. Notie that, in the unphysial ase = 1, Ation (1.15) beomes a

onvexfuntional of (;E). Then, the orresponding hange of sign for log in

equation (1.14) transforms system (1.13), (1.14) into an ellipti system in spae

timevariables(x;s).

2. Harmoni orwave maps up to rearrangement

Letus nowombinethe oneptsofharmoni(or wave)maps andrearrange-

ments.

Definition2.1. Wesaythat

(t;x;a)2UA!X(t;s;a)2D;

isaharmoni map (resp. awave map) upto rearrangement, in shortaHMUR

(resp. aWMUR ),if itisaritial pointof

J(X)=inf

Y Z

UA 1

2 (j

t

X(t;s;a)j 2

+j

s

Y(t;s;a)j 2

)dtdsda;

(2.1)

(4)

withrespettoperturbations withompatsupportinU,wherethe inmumisper-

formedover all possible rearrangement Y of X with respet toa 2A, i.e. for all

funtionsY suhthat

Z

A

h(Y(t;s;a))da= Z

A

h(X(t;s;a))da;

(2.2)

holds true for all h2C(D), (t;s)2U. We all J(X)the Dirihlet integral of X

\up torearrangement"(even inthe ase= 1, forsimpliity).

Our main goal is to show that the equations governing the HUR maps are

related to Fluid Mehanis, and morespeially to the equations of isothermal

irrotationalinvisidgas. Inordertoestablishthislink,itisusefultointroduethe

\phase"densityf(t;s;x;)0assoiatedtoagivenmapX(t;s;a)anddenedby

f(t;s;x;)= Z

Æ(x X(t;s;a))Æ(

t

X(t;s;a))da;

(2.3)

for(t;s;x;)2UDR d

.

Definition2.2. We say that X(t;s;a) is a homogeneous map if the phase

densityf denedby(2.3) ist independentandfatorized

f(s;x;)=(s;x)G();

(2.4)

whereG>0isasmooth radiallysymmetri probability densityonR d

and>0a

probablitydensity on[0;S℄D.

Oneof ourmain result(Theorem5.1) will showapreise orrespondene be-

tweensmoothsolutionsofisothermal irrotationalinvisidgas dynamisequations

andsmoothHUR mapswhiharehomogeneousin thesenseof Denition2.2. In

addition, it will be shown that for mostof suh maps, the density G must be a

enteredGaussian lawwithuniformtemperature>0,namely

G()=(2) d=2

exp(

jj 2

2 ):

(2.5)

Amodel ofolletively vibrating strings. Theoriginalmotivationofthe

oneptofharmoniandwavemapsuptorearrangementwasthefollowing(arti-

ial,uptoourknowledge)mehanialmodelofvibratingstrings,whihorresponds

tothease = 1. Letusonsideraolletionofvibratingstringsparameterized

by

s2[0;S℄!X(t;s;a)

and labelled by a 2 A. The kineti energy of eah vibrating string is evaluated

individually

Z

S

0 1

2 j

t

X(t;s;a)j 2

ds

and thenintegrated in a2Aand t 2[0;T℄. This leadsto therstpartof J(X),

dened by(2.1). ThepotentialenergyofX isnotevaluatedindividuallyforeah

vibratingstringbytheusualformula

Z

S

1

2 j

s

X(t;s;a)j 2

ds;

(5)

butratherolletivelybyrearrangingthelabelsofthestringsforeahxedvalue

of(t;s),in ordertogetthelowestpossibleenergy. Inotherwords,weonsiderall

possibleolletions oftitiousvibratingstrings

s2[0;S℄!Y(t;s;a)

havingthesamespaerepartitionasX, i.e. satisfying(2.2),and omputein this

lasstheinmumof

Z

A Z

S

0 1

2 j

s

Y(t;s;a)j 2

dsda:

Integrating in t the resulting inmum leads to the seond part of J(X). As will

be shown, a WUR map is homogeneous, in the sense of denition 2.2, as the

strings vibrate with a olletive Gaussian distribution of veloities with uniform

temperature>0. A ruderversionofthis modelof vibratingstringshas been

onsideredin[Br2℄,wherethevariablesisdisretewithtwovalues,say0and1and

thepotentialenergyisdenedwithanitediereneinsteadofansderivative. The

orrespondingmodelislinkedtoinompressibleFluidMehanis(Eulerequations)

andPlasmaPhysis(Vlasov-Poissonequations).

3. Evaluationof the Dirihlet integral \up to rearrangement"

Our rst result gives amore tratable formula for the Dirihlet integral \up

torearrangement",involvingaseond-orderlinearelliptiequation,atleastinthe

asewhenthelawofX isasmoothfuntion >0.

Theorem 3.1. LetX(t;s;a)be amap UA!D withD=T d

. Assumeits

law

(t;s;x)= Z

Æ(x X(t;s;a))da (3.1)

tobeasmooth (stritly)positive funtion on UD. Then

J(X)= Z

UA 1

2 (j

t

X(t;s;a)j 2

+j

s

Y(t;s;a)j 2

)dtdsda (3.2)

whereY is denedby

s

Y(t;s;a)=(r)(t;s;Y(t;s;a)); Y(t;s=0;a)=X(t;0;a);

(3.3)

and for eah xed t and s, (t;s;x) is the unique solution, with zero mean, in

x2T d

,ofthe ellipti equation

r:(r)=

s : (3.4)

Aproofisprovidedinsetion8.1.

4. The HMUR and WMUR equations

OurseondresultprovidestheequationssatisedbysmoothHUR(orWUR )

mapsX withsmoothlaw>0.

Theorem 4.1. Let X(t;s;a) be a smooth map U A ! D with D = T d

.

Assumeitslaw

(t;s;x)= Z

Æ(x X(t;s;a))da (4.1)

(6)

to be a smooth (stritly) positive funtion on U D. Then X is HUR (resp.

WUR ) if and only if there are two smooth real funtions and q depending on

(t;s;x)2UD suhthat

s

+r:(r)=0;

s +

1

2 jrj

2

=q;

(4.2)

tt

X(t;s;a)=( rq)(t;s;X(t;s;a)); (t;s;x)= Z

Æ(x X(t;s;a))da;

(4.3)

with = 1 (resp. = 1). In addition, (4.3) an be expressed in terms of the

phasedensityf denedby (2.3)as:

t

f+:r

x

f rq:r

f =0;

Z

f(t;s;x;d)=(t;s;x):

(4.4)

The proof of Theorem 4.1 is provided in setion 8.2. Subsequently, we all

HMUR (resp. WMUR ) equations the ombination of (4.2) and either (4.3) or

(4.4),for =1(resp. = 1). Thestudy oftheinitial valueproblemis entirely

open. Theouplingbetweenequations(4.2)and(4.4)isverypeuliar-bothtand

splaytheroleofa\time"variable,tfor(4.4)andsfor(4.2)-and,ofourse,highly

nonlinear. However,inthespeialaseof homogeneoussolutions,in thesense of

Denition 2.2,theseequationsanberelatedtolassialFluidMehanis.

5. Linkbetween HMUR equations and gasdynamis

Theorem 5.1. Letusonsiderasmoothsolution(f;;)totheHMUR(resp.

WMUR ) equations (4.4), (4.2), where f isthe phasedensity dened by (2.3)for

somemap X.

Assumethat does not depend on t and that,for some s

0

2℄0;S[, thereis a non

degenerate maximum pointx

0

,i.e. ( 2

xixj (s

0

;x

0

))>0inthe sense of symmetri

matries.

Then X isahomogeneous mapin the senseof Denition2.2, if andonly if

f(s;x;)=(s;x)G();

where i)G isa enteredGaussian density(2.5) with uniform temperature >0,

ii) (;) does not depend on t and solve the isothermal irrotational invisid gas

dynamis equations (1.13),(1.14) with=1(resp. = 1).

Proof. Notierstthatthenondegenerayonditionisnotartiial. Indeed

(s;x) =1,(s;x) =0, f(s;x;)=G() isatrivialsolutionfor any hoie ofG,

evenwhenGisnotGaussian. Theproofisstraightforward. From(4.4),weget

G():r(s;x) rq(t;s;x)(s;x):rG()

andtherefore,sineandGare(stritly)positive,

:rlog(s;x) rq(t;s;x):rlogG()=0:

Sine G is assumed to be radiallysymmetri, rlogG() is proportionalto . It

followsrstthat

rlog(s;x)= rlogq(t;s;x);

(7)

forsomeonstant,whihannotbeequaltozero,beauseofthenondegeneray

ondition( 2

x

i x

j (s

0

;x

0

))>0whihimpliesthattherangeofr(s

0

;x)spansR d

.

Thus,wehave

q(t;s;x)= log(s;x);

(5.1)

upto anirrelevantadditiveonstant,for someonstant distintfrom 0. Next,

weget

( rlogG()):rlog(s;x)=0:

Sinetherangeofx!r(s

0

;x)spansR d

,weobtain

rlogG()=0;

anddeduethatGistheenteredGaussiandensity(2.5)withuniformtemperature

>0. Finally,using (4.2),weonludethat aneessaryandsuÆientondition

forasolutionX to theHMUR (resp. WMUR )equations tosatisfy(2.4)is that

doesnotdependontandsolvetheisothermalirrotationalinvisidgasdynamis

equations(1.13),(1.14)with=1(resp. = 1).

Remark. Somewhatsurprisingly,thephysialisothermalgasdynamisequa-

tions,with=1,arelinkedto HUR maps,nottoWURmaps. TheWUR maps

orrespond to isothermalgas dynamis equationswith = 1, in whih asethe

systemisunphysialandspae-timeellipti(withimaginaryspeedsofpropagation).

6. MinimizingHUR maps

In the harmoni ase = 1, we an provide a suffiient ondition for a

smooth map X to be a minimizing HUR map. In addition, we an relate the

orresponding minimization problem, whih is ertainly not onvex, to a linear

maximizationproblemwithonvexonstraintsthatanbeseenasitsdualproblem.

Theorem 6.1. Let (X;;;q) be a smooth solution to the HMUR equations

(4.2), (4.3), asinTheorem4.1, with=1andU =℄0;T[℄0;S[. Letusdenote by

X

jU

the restritionof X for (t;x)2U,a2A: Assume

4T 2

supf d

X

i=1;j=1

2

x

i x

j

q(t;s;x)

i

j

j (t;s;x)2UD; jj1g <

2

: (6.1)

Then,X isaminimizingHURmaponUA. Inaddition,theDirihletintegralof

X \uptorearrangement"J(X),denedby(2.1)isequaltothefollowingfuntional

J(X

jU )=sup

Z

UA

((t;s;X(t;s;a);a)n

t (t;s) (6.2)

+(t;s;X(t;s;a))n

s

(t;s))dH 1

(t;s)da

wheredH 1

isthe Hausdormeasurealong U,n=(n

t

;n

s

)(t;s)the externalnor-

mal for (t;s) 2U, and and are smooth funtions depending respetively on

(t;s;x;a)and(t;s;x) subjettothe pointwiseinequality

t +

1

2 jrj

2

+

s +

1

2 jrj

2

0:

(6.3)

Aproofwillbeprovidedin setion8.3.

(8)

7. Generationof minimizingHUR maps from isothermal gasdynamis

Finally,weanprovidearatherexpliitonstrutionofminimizingHURmaps,

whih are homogeneous in the sense of Denition 2.2, from smooth solutions to

the isothermal gas dynamis equations. Subsequently, = 1 is dropped in the

notations.

Theorem 7.1. Let (s;x) >0;(s;x) be a smooth solution of the isothermal

irrotational invisid gasdynamis equations (1.13), (1.14) for s 2[0;S℄ and x 2

D=T d

,withnormalizedtotalmass

Z

D

(s;x)dx=1:

(7.1)

For eah s, let y 2 D ! X

0

(s;y) 2 D a dieomorphism hosen (aording to

Moser'slemma1.1) withlaw (s;x)dx

Z

D

Æ(x X

0

(s;y))dy=(s;x):

(7.2)

Dene A = D [0;1℄

d+1

and solve, for eah xed s and a = (y;w) 2 A, the

autonomous seond-orderODE int2[0;T℄

tt

X(t;s;a)=( r

)(s;X(t;s;a));

(7.3)

withinitial onditions

X(t=0;s;a)=X

0

(s;y);

t

X(t=0;s;a)=V

0 (w);

(7.4)

wherea=(y;w) and

V

0

(w)=(os(2w

1

);::::;os(2w

d ))

p

2log(w

d+1 ) : (7.5)

Assume

4T 2

supf d

X

i=1;j=1

2

xixj

( log(s;x))

i

j

j0sS; x2D; jj1g <

2

: (7.6)

Then, X is a minimizing harmoni map up to rearrangement on U A, where

U =℄0;T[℄0;S[. In addition, X ishomogeneous inthe sense ofdenition 2.2.

Thisresultalsoestablishesanintriguingrelationshipbetween,ononeside,the

isothermal irrotationalinvisidgas dynamisequations, whih arehyperboliand

forwhih theationprinipledoesnotorrespond toaminimization problem (as

explainedinsetion1.3), and,ontheother side,theminimizationoftheDirihlet

integraluptorearrangementforpresribedDirihletboundarydataand thedual

problem(6.2)whih isalinearmaximizationproblemwithonvexonstraints.

Proof. Let us onsider the phase density f assoiated with X by (2.3). By

onstrutionof(X(0;s;a);

t

X(0;s;a)),through(7.2),(7.4),(7.5),f satises

f(t;s;x;)=(s;x)G(); G()=(2) d=2

exp(

jj 2

2 ) (7.7)

(9)

at time t=0. (Indeed, (s;x) is thelaw ofX

0

(s;y), by onstrutionof X

0 from

Moser'slemma,anditiswell-knownthattheGaussianlawanbewritten

G()= Z

[0;1℄

d+1

Æ( V

0

(w))dw;

(7.8)

whereV

0

isgivenby(7.5).) Beauseof(7.3),f satises

t

f+:r

x f+

r

:r

f =0:

(7.9)

where does not depend on t. Sine (s;x)G() is an obvious solution of this

equation, weonludethat (7.7) istrue forall t 2[0;T℄. Bysetting q= log,

weonludethat(X;;)solvetheHMUR equation(4.3),(4.2). FromTheorem

6.1,wededuethatX isaminimizingHURmapandfromTheorem5.1thatX is

homogeneousinthesenseof Denition2.2. Theproofof Theorem7.1istherefore

omplete.

8. Proofs

8.1. Proof ofTheorem3.1. Letusintrodue

j()=inf

Y Z

1

2 j

s

Y(t;s;a)j 2

dadtds;

(8.1)

whereY issubjetto

Z

Æ(x Y(t;s;a))da=(t;s;x);

(8.2)

andissupposedto besmoothand(stritly)positive. Letus prove

j()= Z

UD 1

2

(t;s;x)jr(t;s;x)j 2

dtdsdx;

(8.3)

where solves(3.4). Notiethat theminimization problem involvedin denition

(8.1)requiresnoboundaryonditions. Introdue

E(t;s;x)= Z

s

Y(t;s;a)Æ(x Y(t;s;a))da:

(8.4)

Observe that this vetor-valued measure E is absolutely ontinuous with respet

to . By Jensen's inequality, its (vetor-valued) density, denoted by e(t;s;x), is

squareintegrablewith

Z

1

2

je(t;s;x)j 2

(t;s;x)dtdsdx Z

1

2 j

s

Y(t;s;a)j 2

dadtds:

(8.5)

Inaddition,from(8.4),wededue

s

+r:(e)=0 (8.6)

(Indeed, forallt 2[0;T℄and allsmoothfuntions g(s;x) ompatlysupportedin

0<s<S :

Z

UD (

s

+e:r)dsdx= Z

UD (

s

+E:r)dsdx

= Z

(

s g+

s

Y(t;s;a):rg)(s;Y(t;s;a))dsda= Z

d

ds

(g(s;Y(t;s;a)))dsda=0:)

Thus,wegetalowerboundforj()byminimizingtheleft-handsideof(8.5)in e,

subjetto(8.6). Sine>0isxed,theuniquesolutionisgivenbye=r,where

(10)

(t;s;x) is, for eah xedt, s, theuniquesolution,with zeromean, in x2 T of

(3.4). Sowehave

j() Z

1

2 jrj

2

dtdsdx:

(8.7)

Next, letus dene Y(t;s;a)2 D =T d

for(t;s) 2U, a 2A, by(3.3) and hek

thatthelawofY,denoted by(t;~ s;x)anddened by

~

(t;s;x)= Z

Æ(x Y(t;s;a))da

doesnotdierfrom. First,~solvesthe\ontinuity"equation

s

~

+r:(~ r)=0:

(Indeed, forall testfuntion g(s;x), vanishing at s=0ands=S,for allxedt,

wehave

Z

(

s

g+r:rg)~dsdt= Z

(

s g+

s

Y(t;s;a):rg)(s;Y(t;s;a))dsda

= Z

d

ds

(g(s;Y(t;s;a)))dads=0:)

Ats=0,wehaveY(t;s=0;a)=X(t;0;a). Thus(t;~ s;x) =(t;s;x) at s=0,

sine is thelaw of X. By uniqueness of thesmooth solutionsto theontinuity

equation, we dedue that ~= for all s 2 [0;S℄. So, X and Y are equal up to

rearrangementwithlaw. Itfollowsthat

j() Z

1

2 j

s

Y(t;s;a)j 2

dadtds

(bydenition(8.1))

= Z

1

2

jr(t;s;Y(t;s;a)j 2

dadtds

(beauseof(3.3))

= Z

1

2

jr(t;s;x)j 2

(t;s;x)dtdsdx

(sineisthelawofY)

j()

(by(8.7)),whihompletestheproof.

8.2. Proof ofTheorem4.1. . FromTheorem3.1,wealreadyknowthatan

optimalY assoiated withX anbedened by (3.3),where r solvesthelinear

elliptiequation(3.4),withdatadependingon. Next,goingbaktotheoriginal

denition ofHUR(or WUR )maps, aneessaryand suÆientonditionforX to

beaHUR (oraWUR ) mapisthat(X;Y;q)isasaddlepointfortheLagrangian

Z

1

2 (j

t Xj

2

+j

s Yj

2

)dtdsda Z

(q(t;s;X(t;s;a)) q(t;s;Y(t;s;a)))dtdsda;

(8.8)

whereq(t;s;x)istheLagrangemultiplierfor onstraint(2.2). Weget

tt

X(t;s;a)=( rq)(t;s;X(t;s;a));

(8.9)

(11)

ss

Y(t;s;a)=(+rq)(t;s;Y(t;s;a)):

(8.10)

Therefore,r,denoted bye,mustsatisfy

s

e+(e:r)e=rq:

(8.11)

(Indeed, we have, for all t 2 [0;T℄ and all test funtion g(s;x) vanishing about

s=0ands=T,

Z

rqgdsdt= Z

ss

Y(t;s;a)g(s;Y(t;s;a))dsda

(using(8.10)andthefatthat isthelawofY)

= Z

s

Y(t;s;a) d

ds

(g(s;Y(t;s;a)))dsda= Z

(

s

g+r:rg)rdsdx

(beauseof(3.3),andusingagainthatisthelawofY). Thus,e=r satises

s

(e)+r:(ee)=rq

and, therefore, (8.11), sine>0satises (1.13).) Next,byintegration of (8.10)

inx,weget

s +

1

2 jrj

2

=q;

(8.12)

uptotheadditiontoq(t;s;x)ofanirrelevantfuntionof(t;s). So,wehaveobtained

thedesiredequations(4.2),(4.3)forX;;;q. Finally,equation(4.4)forthephase

densityffollowsfrom(4.3)byastandardargument. (Indeed,forallxeds2[0;S℄

andalltestfuntiong(t;x;)vanishingat t=0andt=T,wehave

Z

(

t

g+:r

x

g rq:r

g)fdtdxd

= Z

((

t

g)(t;X(t;s;a);

t

X(t;s;a))+

t

X(t;s;a):(rg)(t;X(t;s;a);

t

X(t;s;a))

(rq)(t;X(t;s;a)):(r

g)(t;X(t;s;a);

t

X(t;s;a)))dtda

= Z

d

dt

(g(t;X(t;s;a);

t

X(t;s;a)))dtda=0;

sine

tt

X(t;s;a) = (rq)(t;s;X(t;s;a)).) The proof of Theorem 4.1 is now

omplete.

8.3. Proof ofTheorem6.1. Weusetwointermediaryresults:

proposition8.1. LetX(t;s;a)beasmoothfuntiononUAvaluedinD=

T d

. Let J(X)be denedby (2.1) as the Dirihlet integral \up to rearrangement"

ofX. ThenJ(X)J(X

jU

)whereJ(X

jU

)isdenedby(6.2)asinTheorem 6.1.

proposition8.2. LetX(t;s;a)asmoothfuntiononUAvaluedinD=T d

,

and satisfying the assumptions of Theorem 6.1, inluding ondition (6.1) on T.

Then

J(X

jU )

1

2 Z

UA (j

t

X(t;s;a)j 2

+j

s

Y(t;s;a)j 2

)dtdsda;

(8.13)

whereJ(X )isdenedby (6.2)andY issomerearrangement ofX ina2A.

(12)

By ombining Propositions 8.1 and 8.2, we see that, under the assumptions

of Proposition 8.2, X is a minimizing harmoni map up to rearrangement and,

therefore,Theorem6.1follows. Inaddition,weget

J(X)=J(X

jU );

(8.14)

whih relates X to the linear maximization problem (6.2) in (;) with onvex

onstraint(6.3).

8.4. ProofofProposition8.1. LetX(t;s;a)andY(t;s;a)tobeadmissible

fordenition(2.1),whihmeansthatX andY havethesamelaw

(t;s;x)= Z

Æ(x X(t;s;a))da= Z

Æ(x Y(t;s;a))da (8.15)

and

Z

(j

t

X(t;s;a))j 2

+j

s

Y(t;s;a))j)dtdsda<+1:

(8.16)

Letusintrodue

(t;s;x;a)=Æ(x X(t;s;a));

(8.17)

Q(t;s;x;a)=

t

X(t;s;a)Æ(x X(t;s;a)) (8.18)

E(t;s;x)= Z

s

Y(t;s;a)Æ(x Y(t;s;a))da:

(8.19)

As in the proof of Theorem3.1, in setion 8.1, E is absolutely ontinuous with

respettoanditsdensity, e(t;s;x)satises

Z

1

2

je(t;s;x)j 2

d(t;s;x) Z

1

2 j

s

Y(t;s;a)j 2

dadtds:

(8.20)

The left-hand side an be written as a onvex funtion of (;e), beause of the

following(elementary)lemma(see[Br1℄):

Lemma 8.1. Let (;m) apair of measureson someompat spae M,respe-

tively valuedin RandR d

. Dene

K(;m)=supf Z

(y)d(y)+(y):dm(y); (y)+ 1

2 j(y)j

2

0; 8y2Mg;

(8.21)

wherethesupremumisperformedoverallontinuousfuntionsonM ,,respe-

tivelyvaluedinRandR d

. ThenK=+1unless isnonnegative, misabsolutely

ontinuouswithrespetto withavetor-valueddensityvwhihissquareintegrable

withrespetto. Inaddition, asK<+1,

K(;m)= 1

2 Z

jv(y)j 2

d(y):

(8.22)

Thusweanwrite(8.20) as:

K(;E) Z

1

2 j

s

Y(t;s;a)j 2

dtdsda:

(8.23)

Moresimply,Qisalsoabsolutelyontinuouswithrespetto and

K(;Q) Z

1

2 j

t

X(t;s;a)j 2

dtdsda:

(8.24)

(13)

(Asamatteroffat,wehaveanequality,butweneednotit.) Letusnowobserve

thatthepairs(;Q)and(;E)satisfythefollowingintegralidentitiesforallsmooth

funtions (t;s;x;a)and(t;s;x):

Z

UDA (

t

d+r:dQ)= (8.25)

Z

UA

(t;s;X(t;s;a);a)n

t

(t;s)dH 1

(t;s)da;

Z

UD (

s

d+r:dE)= Z

UA

(t;s;X(t;s;a))n

s

(t;s)dH 1

(t;s)da (8.26)

(whereweusethat doesnotdepend onaand Y hasthe samelawasX). Also

notiethat(8.15) anbeexpressedby

Z

UD

q(t;s;x)d(t;s;x)= Z

UA

q(t;s;x)d(t;s;x;a) (8.27)

foralltestfuntionq. Combiningalltheseproperties,wenallydeduethat

J(X)J(Xj

U );

wherewedene

J(Xj

U

)=infK(;Q)+K(;E);

with(;Q)and(;E)subjettoonstraints(8.25),(8.26)and(8.27). Notiethat

theseonstraintsonlyinvolvethevaluesofX(t;s;a)for(t;x)2U whihjusties

ournotationforJ. Equivalently,weanwrite

J(Xj

U

)= inf

;Q;;E sup

;;q;;;~;

~

Z

[ (t;~ s;x)

s

(t;s;x)+q(t;s;x))d(t;s;x) (8.28)

+(

~

(t;s;x) r(t;s;x)):dE(t;s;x)℄+ Z

[((t;s;x;a)

t

(t;s;x;a)

q(t;s;x))d(t;s;x;a)+((t;s;x;a) r(t;s;x;a)):dQ(t;s;x;a)℄

Z

UA

[(t;s;X(t;s;a);a)n

t

(t;s)+(t;s;X(t;s;a))n

s

(t;s)℄dH 1

(t;s)da;

where,,,~

~

aresubjetto

(t;s;x;a)+ 1

2

j(t;s;x;a)j 2

0; (t;~ s;x)+ 1

2 j

~

(t;s;x)j 2

0:

(8.29)

Thisisthesaddle-pointformulationofaonvexminimizationproblemwithonvex

onstraints.

ProvingProposition8.1isnowveryeasy. Letusset

=

t

+q; =r; ~=

s

q;

~

=r

and dene q by (8.12) where (;) are assumedto satisfy(6.3). This provides a

boundfrombelowforJ(Xj

U

). ThisboundisexatlyJ(Xj

U

),asdenedby(6.2).

ThustheproofofProposition8.1isomplete.

(14)

Remark. Byusing Rokafellar'sdualitytheorem(asstatedin [Brz ℄,hapter

1,forinstane),weouldshowthat J(Xj

U

)=J(Xj

U

),whihmeansthat there

isno\dualitygap".

8.5. Proof of Proposition 8.2. Let(;;X)to satisfy theassumptions of

Theorem 6.1. Let us rst introdue Y(t;s;a) 2 D = T d

dened or (t;s) 2 U,

a2A, by

s

Y(t;s;a)=(r)(t;s;Y(t;s;a)); Y(t;s=0;a)=X(t;0;a):

(8.30)

AsintheproofofTheorem3.1, thelawofY is . Next,introdue

(t;s;x;a)= inf

fz(t)=xg Z

t

0 (

1

2 jz

0

()j 2

q(;s;z()))d +z(0):

t

X(0;s;a);

(8.31)

where theinmum isperformedoverall smoothpaths 2[0;t℄!z()2D suh

thatz(t)=x. Letusstatetwolemmata

Lemma 8.2. Thefollowing identityistrue(undertheassumptionsofTheorem

6.1)

Z

UA

(t;s;X(t;s;a))n

s

(t;s))dH 1

(t;s)da (8.32)

= Z

1

2 j

s Yj

2

dtdsda+ Z

qdxdtds;

whereq isdenedby (8.12).

Lemma 8.3. Underondition(6.1),for allt2[0;T℄,theinmumindenition

(8.31)isahievedbyauniquepathz(),0 t,uniquesolutionofthetwopoint

mixedboundaryvalue problem (halfDirihlet, half Neumann)

z 00

()=( rq)(;s;z()); z(t)=x; z 0

(0)=

t

X(0;s;a):

(8.33)

LetuspostponetheproofofLemma8.2,Lemma8.3,andontinuetheproofof

Proposition 8.2. Itfollowsfrom lassialtheoryofHamilton-Jaobiequations(see

[Ba℄forinstane)that,underondition(6.1),(t;s;x;a)isasmoothsolution,for

0tT,totheHJequation

t +

1

2 jrj

2

+q=0:

(8.34)

Thus,bydenition(8.12)ofq, andsatisfyonstraint(8.29). Inaddition,using

(8.33) in the speialase x = X(t;s;a), wesee that z() =X(;s;a) beause of

(8.9)and,therefore,

(t;s;X(t;s;a);a)= Z

t

0 (

1

2 j

t

X(;s;a)j 2

q(;s;X(;s;a)))d + (8.35)

+X(0;s;a):

t

X(0;s;a):

Dierentiationg(8.35)withrespetto tgives

t

((t;s;X(t;s;a);a))= 1

2 j

t

X(t;s;a)j 2

q(t;s;X(t;s;a)):

Integratingthisexpressionwithrespetto(t;s;a)overUAleadsto

Z

(t;s;X(t;s;a);a)n

t

(t;s)dH 1

(t;s)da= Z

( 1

2 j

t Xj

2

q(t;s;X))dtdsda:

(15)

= 1

2 j

t Xj

2

dtdsda qdxdtds

(sineisthelawofX). Byaddingthisequalitytoidentity(8.32),wehavenally

obtained

Z

UA

((t;s;X(t;s;a);a)n

t

(t;s)+(t;s;X(t;s;a))n

s

(t;s))dH 1

(t;s)da

= 1

2 Z

(j

t Xj

2

+j

s Yj

2

)dtdsda:

Fromdenition (6.2), we dedue(8.13) and the proof of Proposition 8.2 is om-

pleted.

8.6. Proof ofLemma8.2. Wehave

Z

UA

(t;s;X(t;s;a))n

s

(t;s)dH 1

(t;s)da=

Z

UA

(t;s;Y(t;s;a))n

s

(t;s)dH 1

(t;s)da

(beauseX andY areequaluptorearrangement)

= Z

UA

s

((t;s;Y(t;s;a)))dtdsda

(byintegrationin s)

= Z

UA (

s

(t;s;Y(t;s;a))+

s

Y(t;s;a):r(t;s;Y(t;s;a)))dtdsda

= Z

UA (

s

(t;s;Y(t;s;a))+jr(t;s;Y(t;s;a))j 2

)dtdsda

(beauseof8.30)

= Z

(

s

+jrj 2

)dxdtds

(sineisthelawofY)

= Z

( 1

2 jrj

2

+q)dxdtds

(bydenition(8.12). Wealsohave

Z

1

2 jrj

2

dxdtds= Z

jr(t;s;Y(t;s;a)j 2

dtdsda= Z

j

s

Y(t;s;a)j 2

dtdsda

(using (8.30)). Thus, identity (8.32) follows and the proof of Lemma 8.2 is now

omplete.

(16)

8.7. Proof of Lemma 8.3. Let us x (t;s) 2 [0;T℄[0;S℄, x 2 D = T ,

a2A, anddene

h(z)= Z

t

0 (

1

2 jz

0

()j 2

q(;s;z()))d +z(0):

t

X(0;s;a);

(8.36)

forall path 2[0;t℄!z()2D suh that z(t)=x. Theseond derivativeofh

withrespetto zisgivenby

D 2

h(z)(~z;z)~ = Z

t

0 (j~z

0

()j 2

d

X

i;j=1 (

2

x

i x

j

q)(;s;z())~z

i ()~z

j ())d

This quadratiform is positivedenite, under ondition(6.1), beauseof thefol-

lowing(modied) Poinareinequality

2

Z

t

0 j~z()j

2

d 4t 2

Z

t

0 j~z

0

()j 2

d;

(8.37)

whihholdstrueforallt>0andforallsmoothfuntionz~suhthatz~ 0

(0)=0and

~

z(t)=0. (ThisPoinareinequality anbeestablishedbyusing Fourierseries,the

inequalitybeingsaturatedby ~z()=sin(

2t

).) It follows that ondition(6.1)on

T issuÆienttoenfore that,aslongastT,theinmumin denition (8.31)is

ahievedbyauniquepathz(),0 t,uniquesolutionofthetwopointmixed

boundaryvalueproblem (halfDirihlet,half Neumann)(8.33). Thus, theproofof

Lemma8.3isomplete.

8.8. Appendix : a proof of Moser's lemma. In the ase of the torus,

the proof is very easy. Firstsolve the Laplae equation (withperiodi boundary

onditions)

= 1

onthetorus(byusing Fourierseries,forinstane). Nextdene

v(;x)= r(x)

~ (;x)

; (;~ x)=(1 )+(x);

for 2[0;1℄andx2D,sothat

~

+r:(v)~ =0:

Then,foreaha2D solvetheinitialvalueproblem

Z(;a)=v(;Z(;a)); Z(0;a)=a; 8a2A=D:

Chek that,for eah 2 [0;1℄, Z(t;:) is asmooth dieomorphism ofD with law

~

(;x)dx. ConludebysettingX(a)=Z(1;a).

Aknowledgements. The author thanks TICAM, Austin, and the Erwin

Shrodinger Institut, Vienna, for their hospitality during the preparation of this

work. HealsothanksMihelRaslefor stimulatingdisussionsaboutationprin-

iplesinFluidMehanis.

(17)

Referenes

[Ba℄ G.Barles,Solutionsdevisositedesequationsd'Hamilton-Jaobi, Mathematiquesetappli-

ations17,Springer,1994.

[Br1℄ Y.Brenier, Minimalgeodesis ongroups of volume-preservingmaps,Comm.PureAppl.

Math.52(1999)411-452.

[Br2℄ Y.Brenier, Derivation of the Eulerequations froma ariature of Coulombinteration,

Comm.Math.Phys.212(2000)93-104.

[Brz℄ H.Brezis,Analysefontionnelleappliquee,Masson,Paris,1974.

[DM℄ B.Daorogna,J.Moser,OnapartialdierentialequationinvolvingtheJaobiandetermi-

nant,Ann.Inst.H.PoinarAnal.NonLinaire7(1990)1-26.

[Si℄ T.J.Sideris,Thelifespan ofsmooth solutionsto thethree-dimensionalompressible Euler

equationsandtheinompressiblelimit,IndianaUniv.Math.J.40(1991)535-550.

CNRS,Laboratoire

Dieudonn

e,

Universit

edeNie,Frane(onleavefromUniver-

sit

eParis6)

E-mailaddress: breniermath.unie.fr

Références

Documents relatifs

On consid`ere une particule non charg´ee de spin 1/2 soumise ` a un champ magn´etique uniforme et constant B ~ parall`ele ` a l’axe Oz.. Quels r´esultats peut-on trouver et avec

Le trinôme garde un signe constant, celui

Les droites (I J) et (CG) sont coplanaires et non parallèles, elles sont donc sécantes en un point

[r]

[r]

Henri Poincaré (1854-1912) a proposé au début du 20ème siècle une façon originale d’étudier la répartition des valeurs d’une table numérique en montrant que pour un bon

[r]

Enoncer le th´ eor` eme de Parseval et le th´ eor` eme de